Skip to content



This how-to does not intend to be a network design nor a best practices document. It is about how to deploy a WAN network using AVD.

About PREVIEW keys in the schema

Some of the WAN functionality is still in PREVIEW. When this is the case it is indicated as such in the documentation. For these PREVIEW features, Everything is subject to change, is not supported and may not be complete.

If you have any questions, please leverage the GitHub discussions board


Please familiarize yourself with the Arista WAN terminology before proceeding:

Design points

  • When deploying CV Pathfinder, the assumption is that the deployment is using CVaaS.
  • The intent is to be able to support having the different WAN participating devices in different inventories.
  • Only iBGP is supported as an overlay_routing_protocol.
  • On the AutoVPN Route Reflectors and Pathfinders, a listen range statement is used for BGP to allow for distributed Ansible inventories.
  • VRF default is being configured by default on all WAN devices with a wan_vni of 1. To override this, it is necessary to configure VRF default in a tenant in network_services.
  • Path-group ID 65535 is reserved for the path-group called LAN_HA.

Features in PREVIEW

  • WAN HA is in PREVIEW

    • While HA is in preview, it is required to either enable or disable HA if exactly two WAN routers are in one node group.
    • For HA, the considered interfaces are only the uplink_interfaces in VRF default.
    • It is not yet supported to disable HA on a specific LAN interface on the device, nor is it supported to add HA configuration on a non-uplink interface.
    • HA for AutoVPN is not supported
    • flow_tracking_settings is in PREVIEW as the model will change in the next release.
    • eos_validate_state is being enriched to support new tests for WAN designs. These new tests are added only in the ANTA preview mode.

Known limitations

  • Zones are not configurable for CV Pathfinder. All sites are being configured in a default zone <region_name>-ZONE with ID 1. Hence, in eos_designs, the transit node type is always configured as transit region.
  • All Pathfinders must be able to create a full mesh
  • No IPv6 support
  • Path-group ID is currently required under wan_path_groups until an algorithm is implemented to auto generate IDs.
  • The name of the AVT policies and AVT profiles are configurable in the input variables. The Load Balance policies are named LB-<profile_name> and are not configurable.
  • LAN support is limited to single L2 using uplink_type: lan and eBGP L3 using uplink_type: p2p-vrfs in conjunction of underlay_routing_protocol: ebgp.
  • All the WAN routers must have a common path-group with at least one WAN route server to be able to inject the default control-plane match statement in the VRF default WAN policy.

Future work

  • HA support out of PREVIEW
  • New LAN scenarios (L2 port-channel, HA for L2 lan using VRRP..)
  • HA for AutoVPN
  • WAN Internet exit
  • import path-group functionality
  • Indirect connectivity to pathfinder
  • BGP peerings on WAN L3 interfaces and associated route filtering.
  • Extra STUN server for a given path-group not connected to pathfinder
  • Auto generation of Path-group IDs and other IDs.
  • Proper OSPF-BGP redistribution in VRF default.
  • Support for OSPF subinterfaces for p2p-vrfs.
  • Increase test coverage in eos_validate_state support for AutoVPN and CV-Pathfinder
  • Path selection outlier detection feature


eos_cli_config_gen schema should support all of the required keys to configure a WAN network, whether AutoVPN or Pathfinder except for the most recent features. This means that any missing eos_designs feature should be supported using custom_structured_configuration functionality. If you find any missing functionality, please open an issue on Github.

Getting started with WAN

Global settings


These global settings must be the same for every WAN device participating in the WAN network.

When using multiple inventories, the recommendation is to include these variables from a common source, for instance leveraging the arista.avd.global_vars Ansible plugin.


The following table list the eos_designs top level keys used for WAN and how they should be set:

Key Must be the same for all the WAN routers Comment
wan_mode Two possible modes, autovpn and cv-pathfinder (default)
wan_virtual_topologies to define the Policies and the VRF to policy mappings
wan_path_groups to define the list of path-groups in the network
wan_carriers to define the list of carriers in the network, each carrier is assigned to a path-group
wan_ipsec_profiles to define the shared key for the Control Plane and Data Plane IPSec profiles.
cv_pathfinder_regions to define the Region/Zone/Site hierarchy, not required for AutoVPN.
tenants the default tenant key from network_services or any other key for tenant that would hold some WAN VRF informaiton
wan_stun_dtls_disable disable dTLS for STUN for instance for lab. (NOT recommended in production)
application_classification to define the specific traffic classification required for the WAN if any.
wan_route_servers Indicate to which WAN route servers the WAN router should connect to. This key is also used to tell every WAN Route Reflectors with which other RRs it should peer with.
ipv4_acls List of IPv4 access-lists to be assigned to WAN interfaces.

Additionally, following keys must be set for the WAN route servers for the connectivity to work:

  • bgp_peer_groups.wan_overlay_peers.listen_range_prefixes: To set the ranges of IP address from which to expect BGP peerings for the WAN. Include the VTEP ranges for all routers connecting to this patfinder.

WAN mode

AVD supports two design types for WAN:

  • AutoVPN
  • CV Pathfinder

By default the mode is set to cv-pathfinder and can be changed using:

wan_mode: autovpn | cv-pathfinder # default: cv-pathfinder

WAN node_types

There are two built-in node types for WAN:

  • wan_router: WAN routers can be AutoVPN edges, CV Pathfinder edges or transits.
  • wan_rr: WAN route servers, used for the AutoVPN RRs and CV Pathfinder Pathfinder nodes.

WAN route servers

The AVD model for WAN has been built with the intention that it should be possible to have the different WAN routers in different inventories.

The top level wan_route_servers allow to indicate to which AutoVPN RRs or to which Pathinders node the routers should connect to. The key can be different per region for instance.

 WAN routers with this configuration will establish static peerings towards pf1 and pf2 and the common path-groups.
wan_route_servers: # (1)!
  - hostname: pf1
  - hostname: pf2
  1. A wan_router with this configuration will establish BGP peering to all the wan_route_servers in the list if it has a common path-group. A wan_rr with this configuration will establish BGP peerings to every other wan_route_servers in the list if they have a common path-group.


It is recommended to always have at least 2 WAN route servers for redundancy and at least one WAN route server in each path group.

When the WAN route servers are part of the same inventory, each WAN router in the inventory is able to pick up the required information to generate the configuration from the inventory device. However, if the WAN route servers are in a different inventory, it is then necessary to add some information under the wan_route_servers key entry. AVD will raise an error if any required information to generate the configuration is missing. Note that the configuration under wan_route_servers takes precedence over what AVD could derive from the device if it is in the same inventory.

WAN STUN handling

WAN STUN connections are configured by default authenticated and secured with DTLS by default. A security profile is configured with an hardcoded root certificate and matching a certifcate <profile_name>.crt and key <profile_name>.key:

management security
   ssl profile STUN-DTLS
      tls versions 1.2
      trust certificate aristaDeviceCertProvisionerDefaultRootCA.crt
      certificate STUN-DTLS.crt key STUN-DTLS.key

These values can be overwritten using custom_structured_configuration.

This configuration requires certificates to be distributed on the WAN devices to be able to authenticate themselves:

  • For CV Pathinder deployments, CloudVision will automatically deploy certificates on the devices.
  • For AutoVPN, the certficiates must be generated and deployed to the devices for the STUN connections to work.

Disabling STUN

It is possible to disable STUN using AVD but this implies that the STUN service is exposed with no authentication.

# Use wan_stun_dtls_disable: true to disable STUN on all WAN devices
wan_stun_dtls_disable: true

# Use wan_stun_dtls_profile_name to overwrite the default profile name STUN-DTLS
wan_stun_dtls_profile_name: profileA

CV Pathfinder hierarchy

When deploying CV Pathfinder, it is required to define a WAN hierarchy grouping WAN sites in WAN regions using the top level key cv_pathfinder_regions. Furthermore each WAN router must be assigned to a region and site in the node settings.

An example hierarchy looks like:

  - name: Region_1
    id: 1
    description: AVD Region 1
      - name: Site11
        id: 11
        location: Somewhere
      - name: Site12
        id: 12
        location: Somewhere else
  - name: Region_2
    id: 2
    description: AVD Region 2
      - name: Site11
        id: 11
      - name: Site21
        id: 21


Site IDs and names must be unique per region.

And then for each wan_router:

    - name: EDGE1
      cv_pathfinder_region: Region_1
      cv_pathfinder_site: Site11

For Pathfinders (wan_rr), it is possible to set the cv_pathfinder_region key. It is not rendered in configuration but leveraged in CloudVision for the visualization.

WAN carriers and path-groups

Path-groups are used to define overlays over the various network transports a WAN can use.

In AVD, the list of desired path-groups must be defined the same across all WAN participating routers using the following top level key.

  - name: MPLS
      static_peers: false
      dynamic_peers: false
      interval: 300
    id: 100
  - name: INET
    id: 101
  - name: LTE
    excluded_from_default_policy: true # (1)!
    id: 102
  - name: Equinix
    default_preference: alternate # (2)!
    id: 103
  - name: Satellite
    id: 104
  1. The expectation here is that LTE shouldn’t be present in automatically generated load balance policies.
  2. This controls the preference allocated to the path-group in a load balance policy when no preference is specified.


The IDs are required and it is possible to turn on and off IPSec at the path-group level for dynamic or static peers.

To allow for a better visualization in CVaaS, AVD implements on level of indirection to be able to specify the carrier for the last mile provider. In addition some settings have been made available at the WAN carrier level like trusted.

Each carrier is associated to a path-group and multiple carriers can be assigned to the same path-group (e.g. all the Internet providers can be differentiated at the carrier level but bundled together at the path-group level).

The list of carriers must also be the same across all WAN routers:

  - name: IPS-1
    path_group: INET
  - name: ISP-2
    path_group: INET
  - name: MPLS-SP1
    path_group: MPLS
    trusted: true # (1)!
  - name: MPLS-SP2
    path_group: MPLS
    trusted: true
  - name: LTE-5G-SP
    path_group: LTE
  - name: Satellite-SP
    path_group: Satellite
  1. AVD won’t require an interface ACL on WAN interfaces towards a trusted WAN carrier.

Flow tracking

For scalabilty reasons, flow-tracking is enabled only on Dps1 interface by default. It can be added on WAN and LAN interfaces using custom_structured_configuration.

WAN interfaces

A WAN interface in AVD is defined under the node settings under the l3_interfaces list. To be considered as a WAN interface by AVD, the l3_interface must have the wan_carrier key defined (which will allow to detect the path-group thanks to the carrier to path-group mapping). The wan_circuit_id is optional and used on CVaaS to provide more information in the visualization as well as in the AVD generated interface description. Finally the key connected_to_pathfinder allows to disable the static peering configuration on a given path-group.


The WAN interfaces may be connected directly to the Internet, it is the user’s responsibility to craft the ACLs required to ensure only the expected traffic is accepted and sent.

    bgp_as: 65000
    cv_pathfinder_region: Region_1
    cv_pathfinder_site: Site11
      - name: cv-pathfinder-edge
        id: 1
        uplink_switch_interfaces: [Ethernet1]
          # This is a WAN interface because `wan_carrier` is defined
          - name: Ethernet1 # (1)!
            peer: peer3
            peer_interface: Ethernet42
            wan_carrier: ISP-1
            wan_circuit_id: 666 # (2)!
            ipv4_acl_in: TEST-IPV4-ACL-WITH-IP-FIELDS-IN # (3)!
            ipv4_acl_out: TEST-IPV4-ACL-WITH-IP-FIELDS-OUT # (3)!
            dhcp_accept_default_route: true
            ip_address: dhcp
          - name: Ethernet2
            wan_carrier: MPLS-1
            wan_circuit_id: 10555
              - prefix:
            connected_to_pathfinder: False
          # This is NOT a WAN interface
          - name: Ethernet3
  1. peer and peer_interface are optionals and used for description.
  2. wan_circuit_id is optional and used for description.
  3. Configure IPv4 ACLs in and out for the L3 interface. The access lists must be defined under ipv4_acls top level key.

WAN policies

The policies definition works as follow:

  • The policies are defined under wan_virtual_topologies.policies. For AutoVPN mode, the policies are configured under router path-selection, for CV Pathfinder, they are configured under router adaptive-virtual-topology.
  • A policy is composed of a list of application_virtual_topologies and one default_virtual_topology.
  • The application_virtual_topologies entries and the default_virtual_topology key are used to create the policy match statement, the AVT profile (when wan_mode is CV Pathfinder) and the load balancing policy.
  • The default_virtual_topology is used as the default match in the policy. To prevent configuring it, the drop_unmatched boolean must be set to true otherwise, at least one path-group must be configured or AVD will raise an error.
  • Policies are assigned to VRFs using the list wan_virtual_topologies.vrfs. A policy can be reused in multiple VRFs.
  • If no policy is assigned for the default VRF policy, AVD auto generates one with one default_virtual_topology entry configured to use all available local path-groups.
  • For the policy defined for VRF default (or the auto-generared one), an extra match statement is injected in the policy to match the traffic towards the Pathfinders or AutoVPN RRs, the name of the application-profile is hardcoded as APP-PROFILE-CONTROL-PLANE. A special policy is created by appending -WITH-CP at the end of the targetted policy name.
  • For the policy defined for VRF default (or the auto-generated one), an extra match statement is always injected in the policy to match the traffic towards the Pathfinders or AutoVPN RRs. The name of the injected application-profile is hardcoded as APP-PROFILE-CONTROL-PLANE. A special policy is created by appending -WITH-CP at the end of the targeted policy name.
    - name: PROD  # (1)!
      policy: PROD-AVT-POLICY
      wan_vni: 42
    - name: default # (2)!
      wan_vni: 1
    - name: PROD-AVT-POLICY  # (3)!
      default_virtual_topology: # (4)!
          - names: [INET]
            preference: preferred
          - names: [MPLS]
            preference: alternate
        - application_profile: VOICE # (5)!
          lowest_hop_count: true
            - names: [MPLS]
              preference: preferred
            - names: [INET]
              preference: alternate
          constraints: # (6)!
            delay: 120
            jitter: 20
            loss_rate: 0.3
          id: 2 # (7)!
        - application_profile: VIDEO
            - names: [MPLS, LTE]
              preference: preferred
            - names: [INET]
              preference: alternate
            loss_rate: 42.0
          id: 4
        # This will not be rendered on devices without MPLS
        - application_profile: MPLS-ONLY # (8)!
            - names: [MPLS]
              preference: preferred
          id: 5
  1. Assign the PROD-AVT-POLICY to the PROD VRF, multiple VRFs can use the same policy.
  2. VRF default will use the AVD auto-generated DEFAULT-POLICY as no policy is set.
  3. Define the PROD-AVT-POLICY
  4. default_virtual_topology is used to configure the default match in the policy. In this case, default traffic will use INET path-group first and MPLS as backup.
  5. This list element configures the policy to apply to traffic the VOICE application profile. This block of configuration will configure the Load Balance policy, the match statement in the policy (in router path-selection for AutoVPN or router adaptive-virtual-topology for CV-Pathfinder) and for CV-Pathfinder, the AVT profile. The application profile must be defined under application_classification.application_profiles. The Load Balance policy will favor direct path over multihop path (lowest_hop_count). MPLS is prefered over the INET path-group.
  6. The constraints are applied on the load-balance policy. If the delay, jitter or loss-rate of a given path-selection path exceeds the defined criteria, the path is avoided and the remaining ECMP paths are used or the next path in line meeting the constraints is used.
  7. The id is used as the AVT profile ID in the configuration.
  8. The MPLS-ONLY application profile won’t be rendered on devices not having this path-group.

LAN Designs

VRF default routing

An important design point to keep in mind is that the current CV Pathfinder and AutoVPN solutions require the Dps1 interface to be in VRF default. This implies that all the WAN interfaces will also be in VRF default and there may also be LAN interfaces in VRF default.

The following LAN scenarios are supported:

  • Single Router L3 EBGP LAN
  • Single Router L2 LAN

The following LAN scenarios are in PREVIEW:

  • Dual Router L3 EBGP LAN with HA


  • the Site of Origin (SoO) extended community is configured as : note: site id is unique per zone (only a default zone supported today).
  • the routes redistributed into BGP via the route-map RM-CONN-2-BGP are tagged with the SoO.
  • the Underlay peer group (towards the LAN) is configured with two route-maps reused from existing designed but configured differently
    • one outbound route-map RM-BGP-UNDERLAY-PEERS-OUT:
      • advertised the local routes tagged with the SoO extended community.
      • advertised the routes received from iBGP (WAN) towards the LAN also marked with the SoO community.
    • one inbound route-map RM-BGP-UNDERLAY-PEERS-IN:
      • deny routes received from LAN that already contain the WAN AS in the path.
      • accept routes coming from the LAN and set the SoO extended community on them.
  • For VRF default, there is a requirement to explicitly redistribute the routes for EVPN. The RM-EVPN-EXPORT-VRF-DEFAULT is configured to export the routes tagged with the SoO.

PREVIEW: Changes ahead

This configuration currently in PREVIEW will change in future version of AVD.

for eBGP LAN routing protocol the following is done to enable HA:

  • the uplink interfaces are used as HA interfaces.
  • the subnets of the HA interfaces are redistributed to BGP via the RM-CONN-2-BGP route-map BGP underlay peer group is configured with allowas-in 1 to be able to learn the HA peer uplink interface subnet over the LAN as well as learning WAN routes from other sites (as backup in case all WAN links are lost).
  • the Underlay peer group is configured with two route-maps
    • one inbound route-map RM-UNDERLAY-PEERS-IN
      • Match HA peer’s uplink subnets (not marked) to be able to form HA tunnel (not exported to EVPN).
      • Match HA peer’s originated prefixes, set longer AS path and mark with SoO to export to EVPN. These will be used as backup from other sites to destinations on HA Peer Router in case all WAN connections on Peer are down.
      • Match all WAN routes using AS path and set no-advertise community. This will be used as backup routes to the WAN in case this router looses all WAN connections.
      • Match anything else (LAN prefixes) and mark with the SoO <bgp_as>:<wan_site_id> to export to EVPN.
    • one outbound route-map RM-UNDERLAY-PEERS-OUT
      • allowing local routes marked with SoO (routes/interfaces defined via tenants + router-id)
      • allowing subnets of uplink interfaces.
      • allow all routes learned from iBGP (WAN)
      • Implicitly denying other routes which could be learned from BGP towards a WAN provider or redistributed without marking with SoO.



This configuration is not supported and has not been thoroughly tested. Use at your own risk.

  • Configure underlay_routing_protocol to OSPF for both the WAN router and the uplink router.

The HA tunnel will come up properly today but route redistribution will be missing so it is not usable.

  • the HA interface(s) is(are) the uplink interface(s) which are automatically included in OSPF.



In the current implementation, OSPF on LAN is not supported as there is no redistribution of route from OSPF to BGP and vice-versa implemented.



Not Implemented - will be using VRRP

CloudVision Tags

arista.avd.eos_designs will generate CloudVision Tags that assist CloudVision with visualizing the WAN.

The tags will only be generated when wan_mode is set to cv-pathfinder.

Device Tags

Tag Name Source of information
Region cv_pathfinder_region
Zone <region_name>-ZONE for wan_router
Site cv_pathfinder_site for wan_router
PathfinderSet name of node_group or default PATHFINDERS for wan_rr
Role pathfinder, edge, transit region or transit zone

Interface Tags

Hint Tag Name Source of information
Type lan or wan
Carrier wan_carrier if this is a WAN interface
Circuit wan_circuit_id if this is a WAN interface

Using separate Ansible inventories

As described in the design principles, the goal is to be able to distribute the WAN routers in separate Ansible inventories.

When leveraging multiple inventories, the arista.avd collection provide capabilites to create global variables. The following example will be leveraging this capability to share required WAN variables accross multiple inventories.

This example contains contains two sites, SITE1 and SITE2 and a dedicate inventory for pathfinder nodes. This would still relevant for configuring AutoVPN in seperate ansible inventories.

Inventory layout:

  • The PATHFINDERS inventory contains the wan_rr node types.
  • SITE1 and SITE2 inventories contains the following node types: wan_router, spine, l3leaf.
Example layout for Ansible multi inventory
├── global_vars
   ├── common.yml
   └── cv_pathfinder.yml
   ├── group_vars
      └── Pathfinders.yml
   └── inventory.yml
├── SITE1
   ├── group_vars
      ├── LEAFS.yml
      ├── SITE1.yml
      ├── SPINES.yml
      └── WAN_ROUTERS.yml
   ├── host_vars
      └── inventory_hostname.yml
   └── inventory.yml
├── SITE2
   ├── group_vars
      ├── LEAFS.yml
      ├── SITE2.yml
      ├── SPINES.yml
      └── WAN_ROUTERS.yml
   ├── host_vars
      └── inventory_hostname.yml
   └── inventory.yml
wan_mode: cv-pathfinder

# When Pathfinders are in a seperate inventory in addition to the hostname you also need to capture the `vtep_ip` and `path_groups`.
  - hostname: pf1
      - name: mpls
          - name: Ethernet1
      - name: internet
          - name: Ethernet2
  - hostname: pf2
      - name: mpls
          - name: Ethernet1
      - name: internet
          - name: Ethernet2

# Suggested variables to share in your global variables accross inventories.

  - name: Global
    id: 1
      - name: SITE1
        id: 101
      - name: SITE2
        id: 102

    password: "htm4AZe9mIQOO1uiMuGgYQ=="

    shared_key: 0110100A480E0A0E231D1E
    shared_key: 0110100A480E0A0E231D1E

  - name: mpls
    id: 101
  - name: internet
    id: 102

  - name: isp-1
    path_group: internet
  - name: mpls-sp-1
    path_group: mpls

    - name: default
      policy: DEFAULT-AVT-POLICY
      wan_vni: 1
    - name: data
      policy: DATA-AVT-POLICY
      wan_vni: 20
    - name: guest
      policy: GUEST-AVT-POLICY
      wan_vni: 10
          - names: [internet]
            preference: preferred
    - name: GUEST-AVT-POLICY
          - names: [internet]
            preference: preferred
    - name: DATA-AVT-POLICY
          - names: [mpls]
            preference: preferred
          - names: [internet]
            preference: alternate